Applications are not split across several cores, the software has to emplore several threads that can get distributed across the cores - OS overhead is low, but if thread interdependance grows it can become a real problem.
The cost of a die is largely proportional to the area of the die (growing a bit faster due to lower yields with larger area dice) - that's why currently Intel's 45nm CPUs can easily beat AMD's 65nm CPU range in respect to pricing.
